Changes M. Seeger, 12/1

I made some "changes" you might not agree with. These are commented out
(BEGIN/END SUGGESTION), and the old version is still in place (BEGIN/END
OLD). We should discuss these.

I like John's idea about a simpler title, but I didn't change the old one.

In 2.2: Is "decided" correct? I'd write "certain". Didnt change it though.

I changed all the a_i into q_i, simply to get the paper consistent for now.
If you don't like that, change it back, but keep it consistent (at least
the theorems).

Added "Note that $c(x) = y$ iff $t(x,y) > 0$.", end of 2.2

Removed "For any set S...", beg. of 2.3. This is already explained in 2.1,
where I added two sentences.

Changed several \pr{..}(..) to \pr{..}[..]

End of 2.3: Added "for finite ${\cal H}$"

Added footnote in 3: Clarifies what we mean by "averages over subsets of
H", namely that many coefficients are close to 0 (none of them is allowed
to be 0).

3.1: Wrote the two Hoeffdings into one, using y.

3.1: Had to mention the more general McAllester. But it's short.

3.1: Rewrote the stratification part a bit, it was unclear to me.

General: P -> Pr for Probability

4.2: Changed last paragraph.

4.3: Changed "sequential" notion to batch. h -> h_{BP}

4.2+4.3: Changed f(x) -> c(x). f(x) is the unsigned class.
